Log in
← Agentic Way of Work
Any kind of work

Every new project starts from zero

How to make the knowledge from past projects findable — and actually used — when the next idea shows up.

You have done related work before. The research is in some document, the reasoning in an old thread, the numbers in a deck. But when a new idea shows up, finding what you already know costs more than redoing it — so you redo it, and the pile of past work keeps growing without ever helping.

Why this keeps happening.

Knowledge saved as loose files has no working structure: nothing separates one project's truth from another's, and nothing summarizes what a folder contains well enough to trust without reading everything. Storage is cheap; retrieval with confidence is the expensive part — and that is exactly what unstructured accumulation never provides.

The way of working.

One project, one Context

Each stream of work keeps its own scoped Context — goals, decisions, findings. Knowledge that applies across projects lives one level up, in the Space.

A narrative index per folder

Every folder's index says in prose what is inside and what still matters. An Agent — or you, six months later — reads the index first and loads detail only when needed.

Agents search before they work

Starting a task, the Agent searches existing Context first. A past decision or finding either answers the question or becomes the starting material — automatically, because the discipline says so.

How we run it

When we planned the relaunch communication for our own product, the analysis from the preceding migration project was already in SharedContext — per-user knowledge maps, structured summaries. The plan started from that material in the first session. Nothing was rebuilt, nothing re-explained.

Start here.

Organize one real area of your work into projects with Organizing Context, then let the next session begin from it with a first handoff.

Bring the method to your work.

Connect from Claude, Codex, ChatGPT, or any MCP client.

Get connected